AKSHAR: A mechanism for inputting Indic scripts on ... - Pranav Mistry

Pranav Mistry
MIT Media Laboratory
20 Ames St. E15-320F
Cambridge, MA 02139 USA
pranav@media.mit.edu
Niranjan Nayak
Microsoft India Development
Center
Microsoft Campus, Gachibowli
Hyderabad 500046 India
niranjan@microsoft.com
Copyright is held by HCI Hyderabad
USID2007, June 18 – June 20, 2007, Hyderabad, India
<strong>AKSHAR</strong>: A <strong>mechanism</strong> <strong>for</strong> <strong>inputting</strong>
<strong>Indic</strong> <strong>scripts</strong> on digital devices
Abstract
Text input in Indian languages on digital devices
presents unique challenges to the field of humancomputer
interaction. Part of the complexity arises due
to the structure of the <strong>Indic</strong> <strong>scripts</strong>, which are derived
from the ancient Brahmi script. Mechanisms to input
text in Indian languages on mobile phone have been
around <strong>for</strong> a while but none of these <strong>mechanism</strong>s are
usable enough to emerge as the de-facto standard.
These <strong>mechanism</strong>s fail to give clear justice to the
structure, inherent to the Brahmi script that is radically
different than the Roman and the Latin <strong>scripts</strong>. In this
paper, we present a new text input <strong>mechanism</strong> that is
based on the phonetic structure of the <strong>Indic</strong> <strong>scripts</strong>.
The main goal of the research was to devise a new text
input method that is targeted to the novice users and
that can be learnt quickly, retained and fair speed can
be achieved. Preliminary evaluation tests indicate that
the new proposed method <strong>for</strong> <strong>inputting</strong> text in Indian
languages is very intuitive and easy to use.
Keywords
<strong>Indic</strong> script input <strong>mechanism</strong> <strong>for</strong> digital devices, Brahmi
script input method, text input method <strong>for</strong> mobile
phone.

Introduction
Text entry is one of the most frequent human-computer
interaction tasks. Although great strides have been
made towards speech and handwriting recognition,
typewriting and <strong>inputting</strong> text using key-presses
remains and will likely be the main text entry method in
the near future. Computers, mobile phones and other
digital devices demand text input schemes that can be
learnt quickly and retained to achieve a fair speed.
These <strong>mechanism</strong>s should be easy to use rather than
“hunt and peck”.
Of India’s nearly one thousand million people, a
sizeable amount of over nine hundred million is
currently excluded from participating in the so called
in<strong>for</strong>mation or digital revolution. One of the main
reasons <strong>for</strong> this is the absence of appropriate input
<strong>mechanism</strong> that can deal with in<strong>for</strong>mation in the
languages that majority of the Indians speak. On the
other end, the mobile phone revolution has caught on
in India. The numbers of mobile phones that are
currently available in India hugely outnumber the
number of desktop computers and other computing
devices available in the country. However, majority of
the mobile phones still have English as the input and
interaction language. A small number of mobile phone
manufacturers have come up with the support <strong>for</strong> a
couple of Indian languages in their phones. However,
<strong>inputting</strong> text using one of these <strong>mechanism</strong>s is a real
challenge even <strong>for</strong> the experienced mobile phone users.
Text entry methods <strong>for</strong> Indian languages like Hindi,
Gujarati, Marathi, Punjabi, Bengali and Telugu are not
keyboard friendly and their entry using QWERTY
keyboards is cumbersome and complicated; it involves
the use of multiple shift modes. Furthermore, there are
a number of competing keyboard designs and no
national standard. QWERTY keyboards are not
particularly amenable to accommodate the phonetic,
non-alphabetic script like <strong>Indic</strong>. Many alternate layouts
exist which are mapped on the Roman alphabets of
QWERTY keyboards. All these approaches have fallen
short. The probable reason <strong>for</strong> that is, without any
consideration of the core structural features of <strong>Indic</strong>
<strong>scripts</strong>, these methods have been localized from
<strong>mechanism</strong>s <strong>for</strong> Latin or Roman based <strong>scripts</strong>. The
same problem applies to the input methods <strong>for</strong> mobile
phone and other interactive digital devices.
Localization to <strong>Indic</strong> <strong>scripts</strong> is a non-trivial task due to
the large number of alphabet set and the significant
structural differences as compared to Roman or Latin
based <strong>scripts</strong>.
Hence, there is a great need <strong>for</strong> an appropriate input
<strong>mechanism</strong>/scheme <strong>for</strong> digital devices <strong>for</strong> various
Indian languages. In this paper, <strong>AKSHAR</strong> is proposed
as a generic input <strong>mechanism</strong> <strong>for</strong> <strong>inputting</strong> <strong>Indic</strong> <strong>scripts</strong>
on digital devices like mobile phones, kiosk booths,
WLLs, interactive TVs, and personal computers. Most
of the <strong>scripts</strong> in India have their base in the ancient
Brahmi script. Akshar uses the unique feature –
phonetic grouping of 5 consonants- of Brahmi script as
the central design element. The paper discusses the
basic structure of <strong>Indic</strong> <strong>scripts</strong>, the <strong>AKSHAR</strong> <strong>mechanism</strong>
and its unique features. User evaluation has shown
that in terms of ease of use and learning, this new
<strong>mechanism</strong> is notably better than any other input
methods in use today.
2

figure 1: Brahmi script
alphabets.
<strong>Indic</strong> <strong>scripts</strong> and text input methods
Phonetic basis of Indian languages
India is a multilingual country with 23 officially
recognized languages [2]. A vast majority of these
languages have their orthography derived from the
ancient Brahmi script. These <strong>scripts</strong> are cumulatively
referred to as ‘<strong>Indic</strong> Scripts.’
Panini's phonetic classification [2] of the Indian
alphabets into vowels (V: A,, Aa…) and consonants (C:
k Kh ga Ga…) serves as a common base <strong>for</strong> all Indian
languages. In addition, there are also a few graphical
signs (G), which are used <strong>for</strong> denoting nasal
consonants and nasalization of the vowels. Figure 1
illustrates the basic standalone vowels and consonants
in the ancient Brahmi script. In Brahmi script,
alphabets are phonetically classified into groups of
consonants and vowels. Consonants and vowels
combine together to describe a word in the language.
The shapes of these vowels and consonants could be
different in different Brahmi derived <strong>scripts</strong>, but their
base is still the same. As shown in the Figure 1, there
are 9 basic standalone vowels. The consonants can be
grouped into 6 groups of 5 consonants each and a last
group consisting of 4 consonants.
figure 2: First five consonants in different <strong>Indic</strong> <strong>scripts</strong>.
Figure 2 shows an example of the first five basic
consonants used in the Brahmi script rendered
differently in some other <strong>Indic</strong> <strong>scripts</strong>. The columns in
the figure represent different <strong>Indic</strong> <strong>scripts</strong> while the 5
rows denote the first 5 consonants. A single syllable
used in an Indian language consists of either a lone
vowel optionally followed by a graphical symbol or a
consonantal syllable consisting of a consonant followed
by a vowel and optionally followed by one or more
graphical symbols. It is also possible to have two
consonants in a syllable. This methodology of having
consonants and vowels in a syllable is unique to <strong>scripts</strong>
that are derived from the Brahmi script. Figure 3
shows the consonant-vowel combinations <strong>for</strong> the letter
Ka in different <strong>Indic</strong> <strong>scripts</strong>.
<strong>Indic</strong> <strong>scripts</strong> have a unique structure which necessitates
the user to type multiple keystrokes to enter one
character. We will explain this with respect to
Devnagari, the script used in Hindi, Marathi, Konkani
and Sanskrit languages. Figure 4 shows the Devnagari
alphabet set with frequently used consonants and
vowels. Devnagari has 53 base letters – 34 consonants
and 19 vowels in addition to numbers and punctuation
marks. The last nine of the 34 consonants are also
known as semi-vowels.
Issues with current input <strong>mechanism</strong>s <strong>for</strong> <strong>Indic</strong>
<strong>scripts</strong>
Theoretically, infinite number of Devnagari characters
can be generated by combinations of consonants and
vowels. Actually, a smaller subset of characters is
currently used in Devnagari. But even this small subset
is quite large that makes it impossible to design a
practical input <strong>mechanism</strong> where one key corresponds
to one character. It is necessary to have multiple
3

figure 4: The Devnagari
Varnamala showing the
most frequently used 12
vowels, 34 consonants
and 4 conjuncts.
keystrokes of the base consonants and vowels to
generate most Devnagari characters. However, enough
keys are not available on QWERTY keyboards, even to
represent the base consonants (34), vowels (19), and
the halant; here we require 54 keys against the 26 that
are available. This problem intensifies when we think
of <strong>inputting</strong> Devnagari on devices like mobile phones
where only 9 or 10 keys are available on the keypad to
accommodate all these characters [6].
Non-keyboarded devices such as mobile phones, Kiosk
booths, Interactive TVs, and WLLs have good
<strong>mechanism</strong>s <strong>for</strong> <strong>inputting</strong> English language through a
number-pad. However, using the same <strong>mechanism</strong> to
input text in Indian languages is a non-trivial challenge
because of the huge matrix of consonant-vowel and
consonant-consonant-vowel combinations that are
possible in Indian languages.
The most common method <strong>for</strong> <strong>inputting</strong> English on a
mobile phone is the multitapping (‘abc’, ‘def’, ….) input
<strong>mechanism</strong>. Here the user presses the key 2(‘abc’) on
the number-pad of the mobile thrice in succession (with
as little time gap as possible) to enter the character ‘c’.
If a similar <strong>mechanism</strong> is used <strong>for</strong> <strong>inputting</strong> languages
based on Devnagari script, then it requires having 5
consonants on each number of the keypad and 12
vowels on the last keypad number. Because of the
timeout problem, entering these numbers becomes
really cumbersome, in case if a mistake is made. For
example if a user wants to enter the standalone vowel
ए which comes 7 th in the vowel list, then he needs to
tap a number-key on the keypad 7 times. In addition,
he needs to tap the key fast enough that can avoid the
timeout problem. If by mistake, he taps the key 8
times, then he needs to add 12 more taps so that he
can come back to the correct character again. Slightly
better <strong>mechanism</strong>s <strong>for</strong> entering texts in Indian
languages such as EziText [3], Inscript [4], PenfoPad
[7], and KeyLekh [5] exist today. However, the issues
that these methods try to address are slightly different
and a few of them are meant <strong>for</strong> desktop computers.
<strong>AKSHAR</strong> input <strong>mechanism</strong>
Challenges
Three issues related to Indian languages pose
challenges <strong>for</strong> the design of the input <strong>mechanism</strong> <strong>for</strong>
<strong>Indic</strong> Script:
1. Structural complexities of the <strong>Indic</strong> <strong>scripts</strong>
2. Cognitive styles of writing and typing
3. Large number of characters
One of the major problems confronting the
development of an input <strong>mechanism</strong> <strong>for</strong> mobile phone
which is <strong>Indic</strong> <strong>scripts</strong> compliant is that of the key-pad.
One of the problems is numerical in nature: how to load
a large number of characters onto the 10 key key-pad
and at the same time ensure that texts can be entered
with the least number of key presses as well as
avoiding user-irritation. On the other hand, providing a
highly unfamiliar key-pad based on character
frequencies would mean user resistance, since the user
would have to learn a new alphabet order alien to him.
Theoretically two simple models exist: a simple solution
is to emulate the English keyboard style and load on
the keys the different characters. However this method
of multitapping does not work <strong>for</strong> <strong>Indic</strong> <strong>scripts</strong> which
admit a large number of characters. The user has to
press too many keys to enter simple words. The
second set of methods includes two-key input method,
4

eZiText [3], PenfoPad [7] and Pronunciation-based
method. Though these methods are fast as compared
to the simple multitapping <strong>mechanism</strong>, they fall short
to provide a de-facto input <strong>mechanism</strong> which can take
advantage of the inherent structural features of the
<strong>Indic</strong> <strong>scripts</strong>.
The basic goal of the research was to devise a new
input <strong>mechanism</strong> based on the structural features of
the <strong>Indic</strong> <strong>scripts</strong>. It was aimed to come up with a
generic text input <strong>mechanism</strong> which can be ported to
various digital devices.
Solution details
The basic design concept of the <strong>AKSHAR</strong> is to make use
of the -UP-DOWN-LEFT-RIGHT- arrow keys, available in
mobile phones or devices such as TV remotes.
The <strong>AKSHAR</strong> interface has two main elements:
1. Number keys and UP-DOWN-LEFT-RIGHT arrow
keys of keypad
2. Display Screen (text region and 6X3 grid)
Since <strong>Indic</strong> <strong>scripts</strong> have an intuitive phonetic grouping
system of 5 consonants in a group, we have assigned
each of these groups to a single number-key on the
keypad. The display screen has two main regions: text
region and 6X3 grid. The user is given visual feedback
in the 6x3 grid at the bottom part of the screen. The
UP-DOWN-LEFT-RIGHT arrow keys or the D-pad (also
known as direction pad, joystick, navigation keys) are
used to traverse through the grid.
figure 5: Visual feedback available on the mobile device when
number 2 on the keypad is pressed.
The Figure 5 shows the 6X3 grid and the text region on
the screen of a mobile phone.
figure 6: 6X3 grid representation on keypress 1.
Pressing a number key on the keypad displays the
group of characters, which can be entered using that
key, in the 6X3 grid region. Figure 6 shows the 6X3
grid representation, which shows the 12 standalone
vowels that will be displayed on the screen, on pressing
the key 1 on the keypad.
5

figure 7: 6X3 grid representations on keypress 2 to 8.
Figure 7 shows the 6X3 grid representations that will be
displayed on the screen, on pressing the key 2, key 3,
key 4, key 5, key 6, key 7, and key 8 on the phone
keypad, respectively. Each of these groups is phonetic
group of 5 consonants in the Devnagari script, so it is
very intuitive <strong>for</strong> the user to refer them by the first
character of the group.
When a particular number on the keypad is entered,
the user sees all the possible vowels/consonants that
can be entered using that number. For example, when
the number 2 on the keypad is entered, the screen
provides visual feedback as shown in the Figure 8. By
default the first character (क in this case) of the group,
which is at the center in the grid, is inputted in the text
region. The D-pad can then be used to select the actual
character that is desired from the characters displayed
in the 6X3 grid. For example if the user wants to enter
the letter ख, then he types 2 on the keypad (which
6

figure 8: Visual feedback available on the mobile device when
number 2 on the keypad is pressed.
figure 9: Visual feedback available on the mobile device when
number 2 on the keypad is pressed, followed by the up arrow
key.
shows क and related consonants of the phonetic group)
followed by the up arrow key, which would input ख as
shown in the Figure 9.
figure 10: 6X3 grid representations on keypress 9 the first
time.
The <strong>mechanism</strong> <strong>for</strong> entering vowel matras is also
unique and simple. Figure 10 shows the 6X3 grid that
gets displayed to the user when the number 9 on the
keypad is entered.
figure 11: 6X3 grid representations on keypress 9 the first
time when क (Ka) has been entered as the earlier character.
In case of the vowel matras, slightly more feedback is
provided to the user. The grid shows the rendered
<strong>for</strong>ms of the consonant/consonant combination that
was entered along with the different possible matras
applied to that consonant/consonant combination. For
7

example, on pressing the key 9 on the keypad, the
user will see the grid as displayed in the Figure 11, if
the previously inputted character is क.
figure 12: 6X3 grid representations on keypress 9 the second
time when का (Kaa) has been entered as the earlier character.
Graphical symbols (such as the chandrabindu and the
anuswar) can also be entered using the key 9 on the
keypad. Pressing the key 9 on the keypad the first
time allows the user to select a vowel matra and
pressing the key 9 the second time allows the user to
enter the graphical symbols. For example, on pressing
the key 9 second time on the keypad the user will see
grid as displayed in the Figure 12, if the previously
typed character is का (entered by pressing the key 2
following the key 9 on the keypad).
Some commonly used signs and symbols such as
question mark, comma, semi colon, sentence end and
blank space are mapped on the key 0, key #, and key
* of the keypad as shown in the Figure 13.
figure 13: Signs and symbols mapped with key 0, # and *.
The major advantages of the <strong>AKSHAR</strong> <strong>mechanism</strong> in
comparison to the other exiting input <strong>mechanism</strong>s are:
• No timeout problems, since the user uses the
arrow keys on the device to select the character
(unlike multi-tapping, where the same key has to
be tapped multiple times in order to select a
particular character).
• Visual feedback to the user, by providing the
group of characters, which can be entered using a
key, in the 6X3 grid region.
• Less number of clicks, in order to select a
particular vowel or consonant.
• Clear mental model, since the alphabet grouping
model is the same model that people learn in their
schools.
• Smaller learning curve, since the visual feedback
and intuitive phonetic grouping model makes it
simple to use <strong>for</strong> the novice users.
8

The other important feature of the <strong>AKSHAR</strong>
implementation is the Unicode [9] compliancy. Having
support of the Unicode standards enables it to be a
generic <strong>mechanism</strong> across different plat<strong>for</strong>ms and
devices. The Unicode also enables the users to type a
message or text in multiple languages is another
interesting feature, the <strong>mechanism</strong> provides.
figure 14: Working prototype of the <strong>AKSHAR</strong>.
We implemented a fully working version of <strong>AKSHAR</strong> to
input <strong>Indic</strong> <strong>scripts</strong> on mobile phones. Current version
of <strong>AKSHAR</strong> runs on Microsoft Windows Mobile 5.0
plat<strong>for</strong>m and has been implemented on .NETCF. Figure
14 shows a working prototype of <strong>AKSHAR</strong> on a mobile
phone.
User evaluation and feedback
To evaluate the effectiveness of the <strong>AKSHAR</strong>
<strong>mechanism</strong>, we tested the <strong>mechanism</strong> with several
users of varied profiles. We gave different words and
sentences of varied complexities to the users to type in
using the <strong>AKSHAR</strong>.
The main goal of user evaluation was to evaluate the
effectiveness of the <strong>mechanism</strong> in terms of ease of use,
speed, learning speed and user’s mental model. We
also wanted to have qualitative feedbacks about the
interaction experiences of the first time users and users
who have used the <strong>mechanism</strong> <strong>for</strong> a phase of time.
From our user studies we can conclude that:
• Akshar could achieve its primary design objective –
to devise an easy to use and easy to learn input
<strong>mechanism</strong> <strong>for</strong> Indian languages. Some first-time
users could type with no instructions. All could type
with minimal instructions. Instructions seemed to
be easy to remember – users who typed once could
spontaneously help others.
• The user evaluations demonstrated that first-timers
can achieve acceptable speed of use <strong>for</strong> small
typing tasks. Users also perceived the <strong>mechanism</strong>
to be very intuitive to use on the first glance.
• The task list given <strong>for</strong> the user testing showed that
a motivated user can achieve acceptable speed of
use with a few hours of practice on Akshar.
• The most significant success of <strong>AKSHAR</strong> was the
empowerment and liberation experienced by the
users because they could actually type in their
mother tongue.
User evaluation studies over a time inferred that:
• The system has drastic improvement in the terms
of number of key-presses <strong>for</strong> typing some message
9

in mobile phones, as compared to other current
methods in use.
• The system is also powerful than the two key
<strong>inputting</strong> <strong>mechanism</strong>s, because of visual support of
the interface in selecting the character.
• Unlike other methods, the system has a great
advantage of having no timeout problem <strong>for</strong> words
with adjacent characters which are on the same
key.
• The learning curve <strong>for</strong> <strong>AKSHAR</strong> is very small, as the
model of grouping of 5 consonants is the same as
that of the users’ mental model <strong>for</strong> remembering
<strong>Indic</strong> script alphabets.
• The interaction is very intuitive even <strong>for</strong> first time
users.
In overall, the new <strong>mechanism</strong> is very efficient as
compared to the existing inputs methods in use <strong>for</strong>
<strong>inputting</strong> <strong>Indic</strong> <strong>scripts</strong>.
Conclusion and future work
In this paper, we described a <strong>mechanism</strong> that can be
used to input text in Indian languages, which are
derived from the ancient Brahmi script. In the world,
the percentage of people who speak languages that use
<strong>scripts</strong> derived from Brahmi script is also a huge 22
percent. This is also an advantage as the input
<strong>mechanism</strong> we are proposing can be used to input a
large number of <strong>scripts</strong> with some parameterization of
the software. In addition to mobile phones, the scope
of the <strong>AKSHAR</strong> <strong>mechanism</strong> can be extended to other
devices and plat<strong>for</strong>ms like Kiosk booth, Interactive TV,
WLL phones, and also the personal computers.
We are currently conducting a longitudinal study to plot
the learning curve of <strong>AKSHAR</strong> <strong>mechanism</strong> over a period
of time. We plan to compare <strong>AKSHAR</strong> with other
10
current <strong>mechanism</strong>s <strong>for</strong> <strong>inputting</strong> text on digital devices
like KIOSK, WLLs and Interactive TVs. Pilot
deployments of <strong>AKSHAR</strong> have been made in several
locations in urban and rural areas. We hope to
commercialize <strong>AKSHAR</strong> soon and get response from the
market.
Currently, <strong>AKSHAR</strong> supports Gujarati, Hindi, Marathi,
Punjabi, Telugu, Kannada and Bengali as input
languages. We are working on the next prototype
version of <strong>AKSHAR</strong> that supports other Indian
languages, too. In the next prototype, we are
exploring the possibilities of combining the <strong>AKSHAR</strong>
<strong>mechanism</strong> with other powerful input <strong>mechanism</strong>s like
T9 [8] in order to make it even more powerful. On
other end, we also plan to support word-prediction and
dictionary support.
Acknowledgements
We wish to acknowledge the special contributions of
Prof Anirudha Joshi, <strong>for</strong> his inputs in the design of the
<strong>AKSHAR</strong> <strong>mechanism</strong>. We also thank Abhijeet Rokade,
Abhishek Thakkar, Amisha Banker of Industrial Design
Center, IIT Bombay <strong>for</strong> their valuable inputs and
support during the project. We are grateful to Mr. Srini
Koppplu, Mukund B., Pritkamal Singh and Vijayshree
Choudhary of Microsoft India Development Center <strong>for</strong>
their support in the implementation of the working
prototype of the <strong>AKSHAR</strong>.
References
[1] C-DAC multilingual technologies [Online]. Available
from: http://www.cdac.in/html/mlingual.asp [Accessed
5th March 2006].

[2] Department of Official Languages [Online].
Available from: http://rajbhasha.nic.in [Accessed 12th
January 2006].
[3] eZiText [Online]. Available from:
http://www.zicorp.com/eZiText.htm [Accessed 10th
January 2006].
[4] Inscript keyboard layouts [Online]. Available from:
http://www.mit.gov.in/tdil/keyoverlay.htm [Accessed
10th January 2006].
[5] Joshi, A., Ganu, A., Chand , A., Parmar, V.
and Mathur, G. Keylekh: a keyboard <strong>for</strong> text
entry in indic <strong>scripts</strong>, In CHI '04 extended
abstracts on Human factors in computing
systems, ACM Press (2004).
11
[6] MacKenzie, I. S., and Soukoreff, R. W. Text entry
<strong>for</strong> mobile computing: models and methods, theory and
practice. Human-Computer Interaction. 17(2002), 147-
198.
[7] PenfoPad [Online]. Available from:
http://www.penfosys.com/penfopad.htm [Accessed
12th January 2006].
[8] T9 Text Input [Online]. Available from:
http://www.t9.com [Accessed 23rd February 2006].
[9] Unicode standard <strong>for</strong> <strong>Indic</strong> <strong>scripts</strong> and languages
[Online]. Available from:
http://unicode.org/faq/indic.html [Accessed 19th
January 2006].